Fiscal Policy over the Election Cycle in Low-

Income Countries

Christian Ebeke and Dilan Ölçer

WP/13/153

© 2013 International Monetary Fund WP/13/153

IMF Working Paper

Strategy, Policy, and Review Department

Fiscal Policy over the Election Cycle in Low-Income Countries1

Prepared by Christian Ebeke and Dilan Ölçer

Authorized for distribution by Chris Lane

June 2013

Abstract

Focusing on Low-Income Countries, we investigate the behavior of fiscal variables during and after

elections. The results indicate that during election years, government consumption significantly

increases and leads to higher fiscal deficits. During the two years following elections, the fiscal

adjustment takes the form of increased revenue mobilization in trade taxes and cuts to government

investment, with no significant cuts in government consumption. Using a new dataset on national

fiscal rules and IMF programs, we find that both the presence of fiscal rules and IMF programs help

dampen the magnitude of the political budget cycle in LICs. We conclude that elections not only

imply a macroeconomic cost when they take place but also trigger a painful fiscal adjustment in

which public investment is largely sacrificed.

JEL Classification Numbers: D72; E62; P16; P26

Keywords: Elections, budgets, fiscal rules, IMF programs

Author’s E-Mail Address: Cebeke@imf.org; Dilan.Olcer@sciences-po.org

1 We would like to thank Chris Lane, Christian Mumssen, Catherine Patillo, Noah Ndela Jean-Frederic, Susan Yang, Alejandro

Guerson, colleagues in the Low-Income Countries Division of the Strategy, Policy, and Review Department, at the IMF as well as

participants at the 6th CESifo Workshop on Political Economy and the lunch seminar at Sciences Po (France).

This Working Paper should not be reported as representing the views of the IMF.

The views expressed in this Working Paper are those of the author(s) and do not necessarily

represent those of the IMF or IMF policy. Working Papers describe research in progress by the

author(s) and are published to elicit comments and to further debate.

2

Content Page

Abstract ..................................................................................................................................... 1

I. Introduction ........................................................................................................................... 3

II. How is fiscal policy conducted over the election cycle? Preliminary evidence .................. 5

A. Baseline specification and data ........................................................................................ 5

B. Baseline estimates ............................................................................................................ 8

Composition of expenditures ............................................................................................ 8

Composition of tax revenues ............................................................................................ 8

Overall fiscal balance ........................................................................................................ 9

III. Dealing with the endogeneity of election timing .............................................................. 10

IV. Domestic and international scrutiny ................................................................................. 11

A. Do fiscal rules matter? ................................................................................................... 11

B. IMF program engagement .............................................................................................. 12

V. Concluding remarks ........................................................................................................... 15

References ....................................................................................................................... 16

Appendix ............................................................................................................................. 19

Table A1: List of countries in the sample and number of national elections by country,

1990-2010 ....................................................................................................................... 19

Table A2. Descriptive statistics. LICs sample, 1990–2010. ........................................... 19

Table A3: Correlates of the participation into Fund programs in LICs. ......................... 20

Regression results ............................................................................................................... 21

Table 1. Estimates of the Political Budget Cycle across selected Fiscal Variables in

LICs: 1990–2010. ........................................................................................................... 21

Table 2: Addressing the endogeneity of the election timing in LICs. 1990–2010. ........ 22

Table 3. Do fiscal rules and IMF programs dampen the Political Budget Cycle in LICs?

......................................................................................................................................... 23

3

I. INTRODUCTION

There is growing literature that assesses the detrimental effects of policy volatility on long

term growth and aggregate welfare (Fatas and Mihov, 2003; 2012). One source of policy

volatility might be related to national elections and the incumbent’s incentive to use the

economic policy instruments for re-election purposes. Re-election minded incumbents might

have an incentive to use policy instruments (fiscal and/or monetary policy) in such a way that

during election years, public spending and/or money aggregate increase to satisfy the median

voter despite potential adverse effects on fiscal sustainability and aggregate macroeconomic

stability.2 These cycles appear because of asymmetric information where voters are assumed

to lack full information about the incumbent’s competencies.

Empirical studies on the political business cycle in the 1970s until the 1990s focused almost

entirely on advanced economies and generally do not find a regular statistically significant

evidence of cycles (Alesina et al., 1997; and Drazen, 2001, provide excellent reviews on the

empirical results). However, more recent studies have shown the existence of politically

driven economic cycles in developing countries on government current expenditures, indirect

tax revenues, and budget deficits (Brender and Drazen, 2005; Shi and Svensson, 2006;

Block, 2002; Schuknecht, 2000; Vergne, 2009; Drazen and Eslava, 2010; Ehrhart, 2012), and

monetary aggregates (Fouda, 1997; Block, 2002).

However, several limitations and pending issues remain. First, most of these studies did not

explicitly focus on Low Income Countries (LICs) but rather pooled together developing

countries. Our analysis focuses on LICs because they are particularly vulnerable when

conducting election-related cycles. With weaker institutional capacity and poor transparency

in budgets, these countries face greater risks of conducting election-related fiscal policies. By

depleting their fiscal buffers during election years, LICs further increase their

macroeconomic vulnerability and limit their capabilities of guarding themselves against

external shocks. Therefore, it is important to better understand the composition of the

political budget cycles in this type of counties and consider ways to mitigate related fiscal

policy volatility.

Second, although these papers give insight into what happens to specific variables during

elections, they do not provide any analysis of the composition of the post-election

adjustment. Block (2002) does analyze, in a sample restricted to Sub Saharan Africa, a

number of fiscal and monetary variables during and after elections and conclude that

government spending shifts toward more visible, current expenditures and away from public

investment. However, there are several limitations of his analysis. The period of study is

restricted to 1980 to 1995, although for many countries in Sub-Saharan Africa, elections

were not competitive before 1990, and even during the first half of 1990s.3 Moreover, the

2 See Brender and Drazen (2008) for an analysis of whether fiscal outcomes affect the re election prospects for

incumbents. They find that fiscal deficits do not improve incumbents’ re-election prospects in general. In

developed countries, a deficit even punishes the incumbent. 3 Our paper analyzes the whole period after 1990, takes into account both presidential and parliamentary

elections, distinguishes between endogenously as well as exogenously timed elections, and performs the

empirical analyses on various fiscal outcomes (expenditures, revenues, budget balance).

4

analysis is limited to presidential systems in order to address issues related to endogenously

timed elections, yet it induces an important selection bias.4 Our paper investigates the

behavior of a comprehensive set of fiscal variables during and two years after national

elections. It seeks to shed light on the main form of fiscal expansion during the election-year

and the composition of the fiscal retrenchment (if any) in the subsequent years. We use a

recent dataset, National Elections across Democracy and Autocracy (NELDA) (Hyde and

Marinov, 2012) for the election variable and follow the convention in this literature (e.g. Shi

and Svensson 2006; Brender and Drazen, 2008) by focusing on the highest level of national

elections. Therefore, we only include legislative elections for countries with parliamentary

political systems and executive elections for countries with presidential elections.

Third, the paper explores the efficiency of two main constraints on the ability of the

incumbent to pursue a politically-motivated fiscal impulse in election years. We formally test

whether active national fiscal rules and IMF programs in LICs help dampen the magnitude of

the political budget cycle by limiting the incumbent’s incentives to significantly modify

fiscal policy for re-election purposes. To our best knowledge, there is no empirical work that

examines the effects of fiscal rules and IMF programs on the likelihood, the size, and the

composition of the political budget cycles in LICs. Inspired by the pioneer work by Rose

(2006) in the case of US states, we test the extent to which national fiscal rules matter in

LICs using the recently published IMF dataset on fiscal rules (Schaechter et al., 2012). This

is an interesting question to investigate as experts always express doubts regarding the

effectiveness of these rules in the LICs context.5 The issue of the effect of IMF programs on

the political budget cycle in LICs is also intellectually attractive. With several LICs having

experienced various waves of IMF programs over the past decades, one important question

could be to which extent IMF program conditionalities have constrained the incumbents’

election-year extravagances. We follow the recent work by Hyde and O’Mahony (2010) but

focus on the dampening role of IMF programs on the political budget cycle in LICs.

Fourth, taking advantage of the comprehensive dataset on elections (NELDA) compiled by

Hyde and Marinov (2012), we are able to address the endogeneity of the election timing

within countries. Moreover, we also factor in the self-selection bias in the decision to adopt

fiscal rules or participate in IMF programs. The econometric models used control for several

variables that ensure that the election effects are well identified so that any shift in fiscal

variables associated with a national election must be interpreted as a discretionary fiscal

policy by the incumbent. These control variables include external sources of financing

4 Block (2002) analyses elections that take place at regular times in presidential systems. However, presidential

regimes have endogenously timed elections too, particularly in developing countries (Shi and Svensson, 2006).

Also, countries with presidential regimes have characteristics that systematically distinguish them from

parliamentary regimes (Persson and Tabellini, 2003).

5 Debt rules are the predominant national rules for LICs, possibly reflecting institutional weaknesses that would

complicate, for example the implementation of expenditure rules or cyclically-adjusted budget balance rules.

However, only 4 LICs use national fiscal rules. Other LICs operate under supra-national rules. However, the

enforcement and compliance with supranational fiscal rules has been, at best, mixed, in most EU member states,

WAEMU countries, and in the CEMAC region (Cabezon and Prakash, 2008; Schaechter et al., 2012).

5

(grants and loans), the current year business cycle captured by the real GDP growth rate, the

inflation rate and other covariates.

Our results indicate that during election years, government consumption increases leading to

higher fiscal deficits by about 1 percentage point. During the two years following elections,

the fiscal retrenchment takes the form of increased revenue effort in trade taxes and cuts to

government investment. However, this post-election partial fiscal adjustment is not translated

into reduced envelopes associated with current spending, or enough revenue mobilization

efforts to fully offset the deviation allowed during the election-year. The paper also finds that

national fiscal rules help to mitigate the cycles in government consumption. Results also

uncover that LICs with an active IMF program during a national election experienced a much

lower political budget cycle compared to non-IMF program years.

The remainder of the paper is organized as follows. In Section 2, we set up the main

empirical framework and present the baseline results. In Section 3, we test the robustness of

our baseline results by factoring in the endogeneity of the election timing. Section 4

investigates the role of fiscal rules and IMF programs as mitigating factors. Section 5

concludes.

II. HOW IS FISCAL POLICY CONDUCTED OVER THE ELECTION CYCLE? PRELIMINARY

EVIDENCE

This section presents the general framework and the data used to assess the dynamic of fiscal

variables during and after the occurrence of a national election in LICs. Several fiscal

outcomes (government consumption, public investment, breakdown of tax revenues, and

budget balance as percentage of GDP) are used to assess the magnitude of the shocks on the

budget during and after elections. The section also discusses the baseline econometric results.

A. Baseline specification and data

This paper estimates several dynamic panel equations linking a given fiscal outcome with the

election dummy while controlling for standard determinants of the given fiscal variable. As it

has now become standard in the literature on fiscal policy, dynamic equations are specified to

control for the inertia characterizing fiscal variables over time. We use a panel dataset that

covers 68 LICs over 21 years, 1990-2010. Around 51 of these LICs have had at least one

election.6 A country is classified as Low-Income if it benefits from the IMF Poverty

Reduction and Growth Trust (PRGT) as of 2010. The choice of time period is based on

available data after the democratic reforms that many countries, particularly in Sub-Saharan

Africa, implemented in 1990 with the consequence of making elections more competitive.

The baseline specification is as follow:

6 See Appendix Table A1 for the list of countries. The panel data is unbalanced since some countries have

missing values.

6

tititi

p

ptipiti YELEY ,,1,

2

0

,1,

ΓX [1],

where tiY , is the fiscal outcome in each country i at year t. ti ,X is the vector of country-level

covariates which follow and slightly augment the empirical literature on the determinants of

government consumption, tax revenues and budget deficits in developing countries (Rodrik,

1998; Keen and Lockwood, 2010; Combes and Saadi-Sedik, 2006). More specifically,

models control for variables such as: real GDP growth rate, inflation rate, trade openness,

foreign aid, external debt, natural resource rents, agriculture value added, and fiscal rules. All

the fiscal variables, inflation rate, external debt, and foreign aid are drawn from the IMF

WEO database whereas real per capita GDP growth series have been downloaded from Penn

World Table 7.1 dataset. Natural resource rents and total population data have been drawn

from the World Development Indicators database.7

The main variables of interest are the three election dummies tiELE , , 1, tiELE , and 2, tiELE ,

which take the value 1 in case of a national election and 0, otherwise. From equation [1], the

three coefficients 1 ,

2 , and 3 , measure the percentage point change in the fiscal variable

during, one year after, and two years after a national election, respectively.8 We use the

NELDA dataset (Hyde and Marinov, 2012) for the election variables and follow the

convention in this literature (e.g. Shi and Svensson, 2006; Brender and Drazen, 2008) by

focusing on the highest level of national elections. Therefore, we only include legislative

elections for countries with parliamentary political systems and executive elections for

countries with presidential elections. The binary election indicator, ELE, takes the value 1

depending on the year in the election cycle, and 0 otherwise, as described above. There were

191 national elections during the sample period. Table A1 in the Appendix shows the

distribution of national elections across LICs over the period of analysis.

One important contribution of this paper is to provide a comprehensive view of the existence

of the political budget cycle on various fiscal variables, from expenditures, to revenues, and

to budget balance. We take advantage of the information on fiscal variables available in the

IMF datasets on both expenditure and revenues composition within countries. This effort is

necessary as many LICs have usually been neglected in previous papers due to lack of data.

The IMF dataset we use is a comprehensive source of information on budget composition

based on data collected by economist desks in the field and officially approved by countries.

Regarding government spending, this paper differentiates between current expenditures

7 Descriptive statistics of all the dependent variables are provided in Table A2 in the Appendix.

8 When investigating how elections affect government spending, one critique may be that the political budget

cycles observed are due to the extra cost of running elections, and not necessarily a strategic allocation that is

driven by re-election incentives. This is a fair point, though not a real concern in the case for low- income

countries. These countries are very poor and highly dependent on aid. Most of the expenses related to elections

are also born through aid. Therefore, we investigate the effect of elections on government spending while

controlling for aid as a share of GDP. The fact that political budget cycles are observed even when keeping aid

constant shows that there is a political incentive even if elections were costly to run.

7

(proxied by government final consumption) and public investment to assess the effects of

elections on the composition of spending.

On tax revenues, we also allow for a certain granularity.9 Instead of using only overall tax

revenue as many other papers in this literature do, we decompose tax revenue into three

categories; direct, indirect and trade taxes.10 We see the merit in distinguishing between these

types of taxes since it would make our analysis richer in terms of understanding which taxes

the government will put a particular effort on collecting during the different years of the

election cycle. Ehrhart (2012) bases her analysis on direct and indirect taxes but we believe

that there might be a political economy story behind trade taxes as well. Since trade involves

crossing borders, it is potentially easier for the government to vary tax effort on these specific

locations on the borders (Stotsky and WoldeMariam, 1997).11 By looking at tax revenue

ratios at a more disaggregated level, our paper provides additional insights regarding the shift

in the composition of tax revenue efforts among various types of tax revenues over the

whole eletion cycle. The focus on LICs also constitutes one important difference from

existing papers.

Equation [1] is a dynamic specification and is used given the strong inertia characterizing the

fiscal variables of interest. Government administrations are constrained by budgets, and the

current budget largely determines the next period’s appropriations. Although such inertia has

been argued to provide some stability and predetermines fiscal spending (Schuknecht, 2000),

the presence of lagged dependent variables and the country-specific effects renders the OLS

estimator biased since the lagged dependent variable is correlated with the error term

(Nickell, 1981). In order to deal with this issue, there are two commonly used estimators; the

difference-GMM estimator (Arellano and Bond, 1991) and the System-GMM estimator

(Arellano and Bover, 1995; Blundell and Bond, 1998). In the difference-GMM estimator,

equation [1] is taken in first differences (to remove country fixed effects), and the first

differentiated variables are instrumented by their lagged values in level. However, Arellano

and Bover (1995) and Blundell and Bond (1998) have shown that when the explanatory

variables are persistent over time, the lagged values of variables in level risk to be poor

instruments for variables in first-differences. In order to improve the efficiency, they propose

the System-GMM estimator, which increases the moment conditions. The equation in levels

9 Non-resource tax revenue mobilization is a major challenge in many LICs. While overall tax revenues

correspond to more than 50 percent of GDP in some countries in the sample, others barely manage to collect 1

percent of GDP. The mean overall tax revenue in the sample is 14.8 percent. The largest contribution to

revenues comes from indirect taxes (mean in the sample is 5.6 percent of GDP), followed by direct taxes (4.4

percent of GDP), and trade taxes (3.8 percent of GDP).

10 Indirect taxes, which are broad-based taxes on goods and services, are paid by most citizens and correspond

to 5.6 percent of GDP in our sample. Direct taxes represent taxes on income, profits and capital gains,

correspond to 4.4 per cent of GDP and are mostly paid by corporations since personal income taxes are almost

non-existent. Finally, trade taxes, which correspond to 3.8 percent of GDP, are taxes on trade and international

transactions paid by corporations.

11 For the Indian states, Khemani (2004) provides an analysis with subcategories of commodity taxes; sales,

excise and trade. Data is not available on this level for the 68 low-income countries that we study in this paper.

8

and the equation in differences are combined in a system, and then are estimated with an

extended GMM system which allows for the use of lagged differences and lagged levels of

the explanatory variables as instruments. Hence, the System-GMM estimator controls for

unobserved country-specific effects as well as potential endogeneity of the explanatory

variables. The paper uses the Windmeijer’s (2005) correction of standard errors for finite

sample bias. Two specification tests check the validity of the instruments. The first is the

standard Sargan/Hansen test of over-identifying restrictions. The second test examines the

hypothesis that there is no second-order serial correlation in the first-differenced residuals.

The number of lags of the explanatory variables used as instruments is usually limited to

reduce the ‘over-fitting’ bias (Roodman, 2009).

B. Baseline estimates

We first report the baseline findings for the expenditure side, thereafter for the revenue side,

and finally for the fiscal balance.

Composition of expenditures

Table 1 presents the results for the System-GMM estimator for the various fiscal outcomes.

Column 1 reports that government consumption increases during election year, with no

significant decrease the two years after elections. The coefficient on ELEt is significant and

shows that on average, consumption as a share of GDP increases by 0.8 percentage point

during the election year. The results for government investment is reported in column 2,

which shows that government investment as a share of GDP decreases by almost 0.4

percentage point the year following the election. This result is statistically significant.

Although the sign of the coefficients for ELEt-2 is also negative, it is not significant.

These results indicate that political budget cycles on government expenditures are present in

LICs. More specifically, the governments in LICs tend to increase consumption expenditures

during election years while investments are unchanged. The post-election adjustment takes

the form of decreased government investment. These results confirm previous claims

(Vergne, 2009) that government spending shifts towards more visible consumption during

election years. In addition, we show that the negative effect on government investment

appears with a lag and implies that publicly financed projects, for example in infrastructures,

stagnate the year after elections. From the politicians’ point of view, this is strategic, since a

stagnation of investments during election years would probably have a negative impact on re-

election prospects. Although we do not study it explicitly in this paper, this post-election

stagnation in investments may have serious consequences for economic growth.

[Table 1 about here]

Composition of tax revenues

Column 3 shows that the effort put by the government in collecting overall taxes improves

significantly in the years following an election. A look at the composition of tax revenues

reveals a more detailed picture of the governments’ effort in resource mobilization over the

election cycle. In column 4, where indirect taxes (taxes on goods and services) are reported,

the results do not suggest the existence of an election related cycle. Although the coefficient

9

on the ELEt variable is negative, it is not statistically significant. The coefficients on ELEt-1

and ELEt-2 are not significantly different from zero either. These results contrast with recent

findings by Ehrhart (2012) who finds a significant and negative impact of elections on

indirect taxes using a broader sample of all developing countries. Our results suggest the

opposite, implying that LICs’ tax policy differs from other developing countries over the

election cycle. The differences of results compared with existing studies can be explained by

at least two factors. First, we have decomposed total tax revenues into various components to

get a better granularity and found that at least for LICs, the impact of national elections is

observed in the case of trade tax revenues during the run to rebuild eroded fiscal buffers.

Second, our results provide a more detailed assessment of the impact of elections on

government indirect tax revenues since it does not pool together taxes on goods and services

with trade taxes, an approach which is different from previous papers. Our results show that

within the broad definition of indirect taxes, it is the trade tax revenue ratio which matters

and not the taxes on goods and services.

In addition, the effort put on collecting direct taxes (on income, profits, and capital gains)

does not either vary along the election cycle, as shown in column 5. Column 6 shows the

results for tax revenues on international trade. Econometric estimates indicate that the

government changes its effort in collecting trade taxes during election years. There is a slight

(barely statistically significant) increase in trade taxes-to-GDP of about 0.11 percentage

point of GDP during the election year, an effort which is maintained and strengthened during

during the two post-election years. This explains why total tax revenues increase one and two

years after elections. Our results clearly suggest that LICs tend to partially rebuild the eroded

policy buffers on the revenue side through increased discretionary tax revenue mobilization

on international trade. This may be due to the fact that trade taxes tend to be relatively easier

to collect in LICs, as these countries tend to be more challenged than advanced and emerging

economies in terms of revenue mobilization, particularly in terms of domestic tax revenues.

Overall fiscal balance

The dynamic of the overall fiscal balance throughout the election cycle mirrors the behavior

of the expenditure and revenue variables (column 7). The overall fiscal deficit ratio increases

by about 1 percentage point of GDP during the election year, and this is mainly driven by the

observed increase in government current spending.12 In the post-election years, there is

certainly an attempt to rebuild the eroded fiscal buffers, but it does not appear large and

balanced enough to generate any significant statistical impact. The decline in public

investment and the observed tax revenue increases in the post-election years constitute the

main adjustment package in LICs, but fall short of fully rebuilding the eroded fiscal buffers.

The irreversibility of government current expenditures represents the main factor behind the

protracted pressure exerted by elections on the overall fiscal performace throughout the

years.

12

The magnitude of the deviation in the fiscal balance attributed to elections is similar to previous results by Shi

and Svensson (2006).

10

III. DEALING WITH THE ENDOGENEITY OF ELECTION TIMING

One potential critique of the baseline results above is that we treated the election variables as

exogenous relative to fiscal policy, which may not be the case. Both timing of elections and

fiscal policies could, for example, be influenced by a number of (unobserved) variables

which are not included in the regressions. There may be a bias if, for example, the timing of

the election is strategically chosen by the incumbent politician to coincide with favorable

economic conditions. One way to address this potential bias is to distinguish between

elections whose timing is predetermined relative to current fiscal policies (Shi and Svensson,

2006). Using information provided in the NELDA dataset, we classify an election as

predetermined if the election took place on the date fixed by an established constitution or

procedure. Conversely, the election timing is considered endogenous if the election was early

or late relative to the date it was supposed to be held per established procedure.13

We create new election indicators, ELEPREi,t and ELEENDOi,t to replace ELEi,t. The

variable ELEPREi,t equals 1 in country i and year t when there was a predetermined election,

and 0, otherwise. The variable ELEENDOi,t equals 1 in country i and year t if an election that

was not predetermined took place, and 0 otherwise. The indicators for the post-election were

coded accordingly. Among the 191 elections in our sample, 56.5 percent are classified as

predetermined.14 We re-estimate the baseline regressions with the new election indicators. If

the baseline results are robust, they should also hold for predetermined elections. The revised

model takes the following form:

tititi

p

ptip

p

ptipiti YELEENDOELEPREY ,,1,

2

0

,

2

0

,,

ΓX [2],

The coefficients of interest are p which capture the impact of elections after ruling out the

effects of elections that occurred in an unpredicted schedule compared to the constitutional

calendar.

Table 2 presents the econometric results. They are very similar to the previous ones in terms

of magnitude and impacted fiscal outcomes. Government current expenditures significantly

deviate from their normal level during election years, leading to an increase in the overall

fiscal deficit of about 1.3 percentage point of GDP. The post-election years are characterized

by an effort to partially rebuild fiscal buffers but this comes with a price. Public investment is

reduced by about 0.4 percentage point of GDP. The result that governments increase their

effort in the mobilization of trade tax revenues still holds. Two years after the election, our

estimates indicate a reduction of the fiscal deficit by about 0.5 percentage of GDP.

13

This coding is done using the variable NELDA6 in the NELDA dataset. An established procedure is

interpreted as according to the constitution.

14 Out of the 191 elections, 108 are classified predetermined and 38 endogenous. We were unable to classify 45

elections.

11

[Table 2 about here]

IV. DOMESTIC AND INTERNATIONAL SCRUTINY

There is an ongoing literature which has tried to identify the role played by various

macroeconomic factors on the magnitude of political budget cycles in developing countries.

O’Mahony (2010) examined the role played by openness (globalization). Vergne (2009) and

Faye and Niehaus (2011) considered the role of media and financing variables such as natural

resource rents and official development assistance as potential factors. Recently, Combes,

Ebeke and Maurel (2013) examined the role of migrant remittance inflows on the magbitude

of the political budget cycles in developing countries. This section examines two main

factors not fully analyzed in the LICs context, as means to dampen the electoral fiscal

manipulation. We distinguish between a domestic institutional constraint on fiscal policy and

the participation into a program with the International Monetary Fund (IMF).

A. Do fiscal rules matter?

The political budget cycle may be reduced in presence of national fiscal rules if the rules

prevent the incumbent from fiscal extravagances in the time of national elections. We focus

on national fiscal rules as they are more effective and enforced than supranational rules in the

LICs context. However, the enforcement and compliance with supranational fiscal rules has

been, at best, mixed, in most EU member states, WAEMU countries, and in the CEMAC

region (Cabezon and Prakash, 2008; Schaechter et al., 2012).

One main challenge to isolate the impact of fiscal rules is to address the obvious endogeneity

(self-selection) of the adoption and the stability of these rules. This issue will be addressed in

the empirical specifications. The literature on the role of fiscal rules on the reduction of

political budget cycles is not large. Based on a study on the US states, Rose (2006) shows

that balanced budget rules help dampen politically driven cycles in overall spending, taxes

and deficits. Not surprisingly, Rose (2006) finds that the stricter the rules are, the weaker are

the cycles. Inspired by this study, we test whether national fiscal rules act as a domestic

scrutiny factor which helps dampen political budget cycles in LICs. We have not come

across any article testing the dampening ability of fiscal rules in LICs context. However, it is

important to be prudent when interpreting the results as only few LICs use national fiscal

rules with limited enforcement or compliance.

The econometric model exploits the interaction term between the national election dummy

and a dummy for the presence of a fiscal rule to quantify the dampening impact (if any) of

the presence of a national fiscal rule during election times. Because the adoption and the

presence of a fiscal rule are likely to be non random, we address this issue by using a dummy

variable capturing whether a national fiscal rule has been in place for at least 5 years.

Basically, the strategy consists in interacting the election dummy with the 5-year lag of the

fiscal rule dummy (FR).15 More formally, the specification is the following:16

15

The reader may wonder whether the proposed identification strategy to assess the impact of fiscal rule is the

best available. For example, it could be interesting to proceed with an instrumental variable strategy to tackle

(continued…)

12

titititititiiti YFRELEFRY ,,1,5,2,5,11, ΓX [3],

The magnitude of the political budget cycle on public consumption in absence of a national

fiscal rule is measured by 1 . In presence of a rule, the size of the electoral fiscal

manipulation is captured by 11 . The main hypothesis is that: 01 ; 01 . This

suggests that the political budget cycle is higher when the country lacks a fiscal rule

compared to the case where the country has one.

Estimation results are presented in Table 3. Results indicate that for LICs without national

fiscal rules (the vast majority of countries), the size of the political budget cycle on

government consumption is around 1 percent of GDP. This result is not that different from

the previous estimations performed early in the paper. However, once the election dummy is

interacted with the national fiscal rule dummy, the coefficient turns negative and statistically

significant. However, the significance of the coefficient of the interaction term is low,

suggesting that the strength of the dampening role of national fiscal rules is still low in LICs

possibly due to the lack of enforcement, compliance and limited numbers of LICs using

national numerical fiscal rules . When focusing on the marginal effect of elections in LICs

having adopted national fiscal rules, our coefficient estimates in Table 3 suggest that the size

of the fiscal deviation during an election year is close to 0.13 percentage point of GDP (1-

0.87 = 0.13).

B. IMF program engagement

The paper further tests whether countries engaged in programs with the IMF are less likely to

experience a political budget cycle. In other words, we assess whether IMF programs act as

an international scrutiny mechanism which constraints incumbents to use fiscal policy for

electoral motives. There are several reasons why IMF programs may contribute to reduce the

magnitude of the political budget cycle in LICs. LICs that enter into IMF agreements are

subject to conditionality. One key component of programs’ conditionality is the adoption of

sustainable macroeconomic policies. As a result, if implemented, conditionality constrains

government finances, making it more difficult for governments to engage in expansionary

fiscal policies during elections. This issue has been discussed and tested by Hyde and

O’Mahony (2010) in the context of a large panel of developing countries (94 countries)

mixing LICs with other developing nations. The authors find that IMF scrutiny of the

economy and pressure on governments to maintain a sustainable fiscal policy make pre-

the potential endogeneity of fiscal rules. However, finding such instrumental variable, which needs to be fully

exogenous to fiscal outcomes, is very challenging. Another strategy may also be to pursue a two-step approach

where a selection equation explaining the decision to have a fiscal rule is estimated and used to control for the

self-selection bias in the fiscal equation. However, with such a small number of LICs having national fiscal

rules, it does not seem suitable to perform the two-step approach.

16 Also, we will disregard the post-election dummies used before and concentrate the analysis on the election

year only since the cycles in government consumption has been observed during election year and cycles on

revenues are less robust. Moreover, we do not need to break down the fiscal rule dummy into subcomponents as

national fiscal rules in LICs are primarily dominated by debt rules.

13

electoral manipulation of government balance less likely. This result appears robust to the

treatment of the selection bias characterizing the decision to request a program with the

Fund. Our paper will follow the pioneer work by Hyde and O’Mahony (2010) in the case of a

large sample of countries but will depart from it in several ways which will be outlined

below.

One important issue in this literature is the potential endogeneity of IMF programs with

respect to both elections and macroeconomic outcomes. Scholars have argued that

governments prefer not to be under IMF agreements during elections (Dreher, 2004), and

research has shown that governments are more likely to enter into IMF agreements after

elections (Przeworski and Vreeland 2000). We will explore this issue in details in the first-

stage selection equation estimated to purge the endogeneity of IMF programs with respect to

both election timing and macroeconomic developments.17 Our paper therefore tries to

robustly investigate the effect of IMF programs on the size of the political budget cycle by

focusing only on LICs.18 We also depart from the previous literature as the main dependent

variable is government consumption, the budget item which was found to be strongly

correlated with elections throughout the paper. As we will explain further below, the

selection bias associated with Fund programs has been carefully accounted for using an

improved version of the standard first-stage probit model identifying the correlates of Fund

programs which take into account LICs specifiities.

To assess the effect of IMF programs (IMF), the following model is specified:

tititititititiiti YIMFELEIMFY ,,1,,3,2,,1, ΓX

[4],

where ti ,

is the selection-correction factor associated with the IMF program dummy. More

specifically, the model includes the selection factor besides the IMF dummy so that 1 can

be interpreted with limited risks of selection bias. The magnitude of the political budget cycle

on public consumption in absence of an IMF program is measured by . In presence of an

IMF arrangement, the size of the electoral fiscal manipulation is captured by 1 . The

main hypothesis is that: 0 ; 01 . This suggests that the political budget cycle is higher

when the country is not currently under an arrangement with the IMF compared to the case

where the country is engaged with the IMF.

17

However, the strength of the bias due to the potential link between IMF arrangement and election is

attenuated by one stylized fact. As discussed by Hyde and O’Mahony (2010), the majority of elections in the

developing world are held while countries are already under an IMF agreement. LICs are more likely to have

intensive program engagement due to their prolonged balance of payments needs.

18 This literature has typically used country samples that mix LICs and middle-income economies, which tends

to overlook the distinct characteristics of LICs as well as the distinct nature and objectives of Fund engagement

in these countries. LICs face a number of challenges that differentiate them from other economies such as the

nature of shocks, access to financing, and long term challenges (poverty reduction, infrastructure needs,

institutional and capacity building, …) which typically imply that the type of Fund facilities and their goals are

quite different than other Fund’s financial instruments to emerging or advanced countries.

14

The correction of the self-selection associated with the decision to participate in an IMF

program proceeds as follow. A pooled probit model on the determinants of IMF programs in

LICs over the period 1990-2010 is estimated. Standard determinants of IMF programs

include previous levels of: external reserves, fiscal balance, trade openness, inflation rate. We

also add to this list the size of natural resource rents, and a dummy variable indicating

whether a national election is scheduled in the next year, and the election variable crossed

with an indicator of electoral competitiveness. These two variables are added to the selection

model to capture to what extent LICs are less likely to request IMF programs in the year

prior to national elections, conditional on the degree of competitiveness in the considered

election.19 This is an improvement to the literature having dealt with the selection bias

associated with the decision of requesting a Fund program. Controlling for resource rents, for

the electoral calendar, and for the degree of electoral competitiveness before the year of Fund

programs help factor in some specificities in the LICs context.

Once the probit is estimated on the group of control variables ti ,Z , the selection correction

factor is computed as follows (see Maddala, 1983; Vella and Verbeek, 1999; Keen and

Lockwood, 2010):20

,0,1

,1,

,

,

,

,

,

,

,

ti

ti

ti

ti

ti

ti

ti

IMF

IMF

Z

Z

Z

Z

where and represent the probability density and cumulative density functions of the

standard normal distribution, respectively.

Estimation results are presented in Table 3. Results point to a negative effect of IMF

programs on the magnitude of the political budget cycle on consumption in LICs. In absence

of an IMF program, government consumption deviates by about 1 percentage point of GDP

during national elections whereas the size of the deviation drops to 0.34 percentage point of

GDP in presence of an active IMF program. Results also indicate a positive and significant

effect of the selection factor ti ,

, which suggests that it was crucial and legitimate to

account for the selection bias in the estimates. Our results have indicated that both fiscal rules

19

The selection equation also helps deal with the potential bias which could arise if IMF lending were

significantly higher during months prior to elections. Dreher and Vaubel (2004) found that it is indeed the case.

We rule out this effect by always controlling for official development assistance in the regressions and by

explictly controlling for the electoral calendar and timing in the selection equation. There are thefore limited

risks that our results are fully driven by IMF lending dynamics before and after elections. Moreover, the

direction of this bias would work in lowering the estimated effect of the IMF programs leading to

underestimnated effects instead of inflated effects.

20 Results of the probit selection model identifying the determinants of LICs’ participation into IMF programs

are available in Appendix A3.

15

and IMF programs play an important role in LICs in limiting the propensity of incumbents to

allow large deviations in government consumption during election years as means to

maximize their chance of re-election. Although the results seem appealing, they should be

interpreted with caution. Indeed, the coefficients associated with the interaction terms of

elections crossed with fiscal rules and IMF programs exhibit a low significance, which

suggests that the dampening role is at play but it is not strong enough to generate more

precise estimates. There are several reasons which can be evoked to explain these results.

First, only few LICs (4 LICs to be precise) have adopted active national fiscal rules, an issue

which certainly contributes to reduce the explanatory power of the fiscal dummy in the

model. It could also be that these rules are not sufficiently enforced, exacerbating the

credibility problems faced by these institutional arrangements in many LICs. Second, despite

the fact the selection equation associated with IMF programs explicitly controls for several

covariates, the self-selection bias is always only partially controlled for. In addition, as the

majority of elections in LICs are held while countries are already under an IMF agreement,

this limits the statistical power of the IMF program dummy in dampening the political cycle.

V. CONCLUDING REMARKS

This paper investigates political budget cycles in LICs by analyzing the behavior of the

following variables throughout the election cycle: government consumption, government

investment, composition of tax revenue, and fiscal balance. We find that during election

years, government consumption increases and leads to higher fiscal deficits. During the two

years following elections, fiscal adjustment takes the form of increased revenue effort in

trade taxes and cuts to government investment. We showed that the size of the political

budget cycle is much lower in countries having adopted national fiscal rules or those

participating in IMF programs during the election period.

We analyzed the behavior of these variables throughout the election cycle because the way

the governments decide to manipulate fiscal policy may have implications for future

economic growth. The results in this paper show that elections not only imply a

macroeconomic cost when they take place, but also trigger a painful fiscal adjustment in

which public investment is largely sacrificed and trade put at risk. Although economic

growth was not explicitly studied in this paper, the different policy tools that low-income

countries seem to be using during the political budget cycle indicate a negative effect on

economic growth. One reason is due to the overall volatility in fiscal policy that elections

trigger. The other reason is because trade may be hampered by the post election increased

effort in mobilizing trade taxes. Similarly, the decrease in investment may directly hamper

growth.

This paper used a novel dataset on fiscal rules to highlight that such rules may help to

dampen election-driven cycles in the budget. Although the mere existence of fiscal rules does

not mean that they will be enforced, it may be a first step toward tying the hands of a

politician or government incentivized to conduct political budget cycles. The paper also

showed that IMF programs in LICs have contributed to lowering the magnitude of the

political budget cycle.

16

References

Alesina, A., Roubini, N., Cohen, G.D., 1997. Political cycles and the macroeconomy. the

MIT Press.

Arellano, M., Bond, S., 1991. Some Tests of Specification for Panel Data: Monte Carlo

Evidence and an Application to Employment Equations. Review of Economic Studies

58, 277–297.

Arellano, M., Bover, O., 1995. Another look at the instrumental variable estimation of error-

components models. Journal of Econometrics 68, 29–51.

Block, S.A., 2002. Political business cycles, democratization, and economic reform: the case

of Africa. Journal of Development Economics 67, 205–228.

Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel

data models. Journal of Econometrics 87, 115–143.

Brender, A., Drazen, A., 2005. Political budget cycles in new versus established

democracies. Journal of Monetary Economics 52, 1271–1295.

Brender, A., Drazen, A., 2008. How do budget deficits and economic growth affect

reelection prospects? Evidence from a large panel of countries. The American

Economic Review 98, 2203–2220.

Combes, J.-L., Saadi-Sedik, T., 2006. How does trade openness influence budget deficits in

developing countries? Journal of Development Studies 42, 1401–1416.

Combes, J.-L., Ebeke, C., and Maurel, M. 2013. Remittances prior to an election. CERDI

Etudes et Documents, 2013-07.

Drazen, A., 2001. The political business cycle after 25 years, in: NBER Macroeconomics

Annual 2000, Volume 15. MIT Press, pp. 75–138.

Drazen, A., Eslava, M., 2010. Electoral manipulation via voter-friendly spending: Theory

and evidence. Journal of Development Economics 92, 39–52.

Dreher, A., 2004. The Influence of IMF Programs on the Re-election of Debtor

Governments. Economics & Politics 16, 53–76.

Dreher, A., and Vaubel, R., 2004. Do IMF and IBRD Cause Moral Hazard and Political

Business Cycles? Evidence from Panel Data, Open Economies Review, 15, 5-22.

Ehrhart, H., 2012. Elections and the structure of taxation in developing countries. Public

Choice 1–17.

Fatás, A., Mihov, I., 2003. The Case for Restricting Fiscal Policy Discretion. The Quarterly

Journal of Economics 118, 1419–1447.

17

Fatás, A., Mihov, I., 2012. Policy Volatility, Institutions and Economic Growth. Review of

Economics and Statistics. doi: 10.1162/REST_a_00265

Fouda, S.M., 1997. Political Monetary Cycles and Independence of the Central Bank in a

Monetary Union: an Empirical Test for a BEAC Franc Zone Member Country.

Journal of African Economies 6, 112–131.

Faye, M., and Niehaus, P., 2012. Political Aid Cycles. American Economic Review, 107,

3516-3530.

Hyde, S.D., Marinov, N., 2012. Which Elections Can Be Lost? Political Analysis 20, 191–

210.

Hyde, S.D., O’Mahony, A., 2010. International Scrutiny and Pre-Electoral Fiscal

Manipulation in Developing Countries. The Journal of Politics 72, 690–704.

Keen, M., Lockwood, B., 2010. The value added tax: Its causes and consequences. Journal of

Development Economics 92, 138–151.

Khemani, S., 2004. Political cycles in a developing economy: effect of elections in the Indian

States. Journal of Development Economics 73, 125–154.

Maddala, G.S., 1983. Limited-Dependent and Qualitative Variables in Econometrics,

Econometric Society Monographs, Cambridge University Press.

Nickell, S., 1981. Biases in Dynamic Models with Fixed Effects. Econometrica 49, 1417–

1426.

O’Mahony A., 2010. Engineering Good Times : Fiscal Manipulation in a Global Economy,

British Journal of Political Science, 41, 315-340.

Przeworski, A., Vreeland, J.R., 2000. The effect of IMF programs on economic growth.

Journal of Development Economics 62, 385–421.

Rodrik, D., 1998. Why Do More Open Economies Have Bigger Governments? Journal of

Political Economy 106, 997–1032.

Roodman, D., 2009. A Note on the Theme of Too Many Instruments. Oxford Bulletin of

Economics and Statistics 71, 135–158.

Rose, S., 2006. Do fiscal rules dampen the political business cycle? Public choice 128, 407–

431.

18

Schaechter, A., Kinda, T., Budina, N., and Weber, A., 2012. Fiscal Rules in Response to the

Crisis - Toward the Next-Generation Rules. A New Dataset, IMF Working Papers

12/187, International Monetary Fund.

Schuknecht, L., 2000. Fiscal Policy Cycles and Public Expenditure in Developing Countries.

Public Choice 102, 113–128.

Shi, M., Svensson, J., 2006. Political budget cycles: Do they differ across countries and why?

Journal of Public Economics 90, 1367–1389.

Stotsky, J., G., and WoldeMariam, A., 1997. Tax Effort in Sub-Saharan Africa, IMF

Working Paper 97/107, International Monetary Fund.

Vella, F., Verbeek, M., 1999. Estimating and Interpreting Models With Endogenous

Treatment Effects. Journal of Business & Economic Statistics 17, 473–478.

Vergne, C., 2009. Democracy, elections and allocation of public expenditures in developing

countries. European Journal of Political Economy 25, 63–77.

Windmeijer, F., 2005. A finite sample correction for the variance of linear efficient two-step

GMM estimators. Journal of Econometrics 126, 25–51.

Appendix

Table A1: List of countries in the sample and number of national elections by country, 1990-2010

Afghanistan 2 Gambia 4 Nepal 3

Armenia 5 Georgia 5 Nicaragua 4

Bangladesh 3 Ghana 5 Niger 4

Benin 4 Guinea 4 Nigeria 4

Bolivia 6 Guinea-Bissau 5 Papua New Guinea 4

Burkina Faso 4 Haiti 6 Rwanda 2

Burundi 2 Honduras 5 Senegal 3

Cambodia 3 Kenya 4 Sierra Leone 3

Cameroon 3 Kyrgyz Rep. 5 Solomon Islands 1

Central African Rep. 4 Lesotho 3 Tajikistan 4

Chad 3 Liberia 2 Tanzania 5

Comoros 5 Madagascar 5 Timor-Leste 1

Congo, Dem. Rep. 1 Malawi 4 Togo 4

Congo, Republic 3 Mali 4 Uganda 3

Cote d’Ivoire 4 Mauritania 5 Uzbekistan 3

Djibouti 3 Moldova 8 Yemen 2

Ethiopia 3 Mozambique 4 Zambia 5

Notes: We only include legislative elections for countries with parliamentary political systems and executive elections for countries with presidential elections. Source: National Election Around the World (NELDA) dataset.

Table A2. Descriptive statistics. LICs sample, 1990–2010.

Variable Obs Mean Std. Dev. Min Max

Election dummy 1330 0.13 0.34 0 1

Government consumption ratio 1145 15.20 7.52 1.53 62.17

Public investment ratio 1044 7.37 5.69 0.08 59.85

Total tax revenue ratio 832 14.68 6.51 1.27 58.11

Taxes on goods and services ratio 666 5.47 3.19 0.04 16.96

Direct taxes ratio 697 4.31 3.07 0.01 23.89

Trade taxes ratio 666 3.88 2.87 0.00 14.12

Overall fiscal balance ratio 1004 -2.48 6.74 -72.35 61.83

Real per capit GDP growth 1273 1.23 7.68 -71.24 64.20

Official Development Assistance ratio 1276 13.22 11.74 -2.56 146.89

External PPG debt ratio 1226 85.47 110.30 0.58 2394.86

Trade openness ratio 1179 76.28 37.18 0.19 213.22

ln (100+Inflation) 1279 5.40 0.65 4.61 10.35

National fiscal rule dummy 1330 0.02 0.14 0 1

ln (Total population) 1330 15.16 1.92 11.13 18.86

Total natural resource rents ratio 1330 4.85 12.29 0 105.73

Reserve coverage (in month of imports) 1204 3.50 2.58 0.00 19.75

Political globalization (from The KOF Institute) 1313 46.40 18.46 6.59 90.90

Change in real percapita GDP growth 1266 0.00 10.00 -60.55 132.33

Note: All variables expressed as “ratios” denote nominal values normalized by nominal GDP of each country.

Table A3: Correlates of the participation into Fund programs in LICs.

Dependent variable: Fund program dummy LPM Probit Period: 1990–2010. (1) (2)

International reserve coverage (in months of imports),t-1 0.00496 -0.0425**

[0.0152] [0.0178]

Fiscal balance-to-GDP,t-1 0.00286 -0.00296

[0.00349] [0.00966]

ln(100+Inflation) ,t-1 -0.0494 0.00539

[0.0867] [0.0650]

Official development assistance-to-GDP,t-1 0.00882*** 0.0417***

[0.00261] [0.00547]

ln(Population) ,t-1 -0.0856 0.146***

[0.313] [0.0307]

(Election*Competition) ,t+1 0.0408** 0.163**

[0.0198] [0.0665]

Election dummy,t+1 -0.244* -0.790*

[0.133] [0.410]

Political globalization (KOF Institute index) ,t -0.00115 0.0204***

[0.00382] [0.00325]

Change in real per capita GDP growth ,t -0.00204** -0.00654

[0.000880] [0.00499]

Intercept 2.085 -3.430***

[4.483] [0.532]

Country-fixed effects Yes No Observations 916 916 R-squared 0.043 0.177 Number of countries 63 63

Note: Robust standard errors in brackets. *** p<0.01, ** p<0.05, * p<0.1. LPM: Linear probability model with country-fixed effects. The variable “Competition” is not included additively because of its mechanical perfect colinearity with the election dummy variable.

Regression results

Table 1. Estimates of the Political Budget Cycle across selected Fiscal Variables in LICs: 1990–2010.

(1) (2) (3) (4) (5) (6) (7) G I T TGS TD TT Bal

Election t 0.841

*** -0.194 0.209 -0.075 0.102 0.112

* -1.047

*

[0.258] [0.210] [0.152] [0.061] [0.083] [0.066] [0.542]

Election t-1 -0.059 -0.371** 0.340

* 0.134 0.023 0.173

*** 0.275

[0.223] [0.159] [0.187] [0.084] [0.084] [0.064] [0.330]

Election t-2 -0.049 -0.081 0.290* 0.026 -0.046 0.212

* -0.262

[0.197] [0.208] [0.159] [0.101] [0.061] [0.114] [0.322]

Lagged dependent variable 0.730***

0.813***

0.893***

1.029***

0.831***

0.976***

0.260***

[0.103] [0.089] [0.168] [0.057] [0.105] [0.064] [0.094]

Real per capita GDP growth -0.032 0.027 0.041** 0.033

*** -0.003 0.016

*** 0.084

***

[0.034] [0.023] [0.020] [0.006] [0.010] [0.006] [0.031]

Official development assistance-to-GDP 0.073** 0.038

** -0.001 -0.000 -0.002 0.005 -0.019

[0.037] [0.015] [0.014] [0.003] [0.003] [0.003] [0.018]

External debt-to-GDP -0.002 -0.001 -0.004 -0.000 -0.001 -0.001 -0.005

[0.003] [0.001] [0.004] [0.001] [0.001] [0.001] [0.003]

Trade openness 0.028** 0.011

** 0.013 0.000 0.008

* 0.002 0.014

[0.012] [0.005] [0.012] [0.002] [0.004] [0.001] [0.010]

ln (100+Inflation rate) -0.149 0.419 -0.049 0.005 0.046 -0.164* 0.582

[0.443] [0.335] [0.214] [0.083] [0.088] [0.094] [0.458]

Fiscal rule dummy t-1 1.625***

0.345 0.179 -0.036 0.281 -0.232** -0.400

[0.470] [0.327] [0.541] [0.170] [0.206] [0.093] [1.603]

Natural resource rents-to-GDP -0.003 -0.005 0.010 -0.003

[0.006] [0.005] [0.013] [0.003]

ln (Total population) 0.020 -0.010 0.035 0.004

[0.104] [0.029] [0.023] [0.049]

Intercept 1.999 -2.086 0.760 0.040 -0.556 0.711 -5.941**

[2.348] [1.785] [2.320] [0.590] [0.653] [1.225] [2.385]

N 1234 1140 815 679 705 679 970 No of countries 60 57 56 52 53 52 61 m1:p-value 0.004 0.001 0.003 0.000 0.022 0.000 0.017 m2:p-value 0.516 0.416 0.530 0.350 0.294 0.104 0.260 Hansen OID: p-value: 0.829 0.060 0.031 0.839 0.249 0.540 0.114 No of instruments 17 17 21 21 21 25 23

Note: Standard errors in brackets. G: Government consumption ratio; I: Public investment ratio; T: Total tax revenue ratio; TGS: Tax revenues on goods and services ratio; TD: Tax revenues on income; TT: Trade tax revenues ratio; Bal: Overall fiscal balance ratio. All equations are estimated using the two-step System-GMM with Windmeijer (2005) correction of standard errors.

* p < 0.10,

** p < 0.05,

*** p < 0.01.

Table 2: Addressing the endogeneity of the election timing in LICs. 1990–2010.

(1) (2) (3) (4) (5) (6) (7) G I T TGS TD TT Bal

Predetermined election t 0.758

** -0.231 0.095 -0.074 0.147 0.098

* -1.278

**

[0.327] [0.183] [0.167] [0.089] [0.116] [0.055] [0.618]

Predetermined election t-1 0.224 -0.398** 0.131 0.053 0.093 0.112

* 0.253

[0.305] [0.188] [0.255] [0.108] [0.086] [0.058] [0.438]

Predetermined election t-2 -0.031 -0.159 0.076 -0.020 -0.057 0.171* -0.466

*

[0.218] [0.308] [0.148] [0.108] [0.096] [0.094] [0.278]

N 1131 1043 758 633 657 631 900 No of countries 60 57 56 52 53 52 61 m1:p-value 0.007 0.002 0.004 0.000 0.009 0.001 0.019 m2:p-value 0.653 0.390 0.236 0.269 0.250 0.089 0.210 Hansen OID: p-value: 0.583 0.022 0.031 0.864 0.238 0.705 0.217 No of instruments 20 20 24 24 24 28 20

Note: Windmeijer (2005) corrected standard errors in brackets. G: Government consumption ratio; I: Public investment ratio; T: Total tax revenue ratio; TGS: Tax revenues on goods and services ratio; TD: Tax revenues on income; TT: Trade tax revenues ratio; Bal: Overall fiscal balance ratio. All specifications include the exact control variables as in Table 1. The models also control for the endogeneous election dummies (dated at year t, t-1, and t-2, respectively) which identify whether the election was early or late relative to the date it was supposed to be held per established constitution or procedure. The predetermined election dummies identify elections which took place on the date fixed by an established constitution or procedure. All equations are estimated using the two-step System-GMM with Windmeijer (2005) correction of standard errors.

* p < 0.10,

**

p < 0.05, ***

p < 0.01

Table 3. Do fiscal rules and IMF programs dampen the Political Budget Cycle in LICs?

(1) (2) G G

Election dummy 1.009

*** 0.963

***

[0.331] [0.293]

Election*Lagged fiscal rule dummy -0.870*

[0.511]

Election*IMF program dummy -0.623*

[0.342]

Lagged fiscal rule dummy 1.986***

1.621***

[0.662] [0.376]

IMF program dummy -0.712

[0.444]

(Predicted selection correction factor) 0.621**

[0.253]

Lagged dependent variable 0.695***

0.837***

[0.089] [0.041]

Real GDP growth -0.017 0.004

[0.041] [0.014]

Official development assistance-to-GDP 0.073* 0.045

***

[0.037] [0.017]

External debt-to-GDP -0.003 -0.001

[0.003] [0.003]

Trade openness 0.032***

0.015***

[0.011] [0.005]

ln (100+Inflation) -1.189 -0.442

[0.976] [0.344]

Intercept 7.809 3.488*

[5.321] [1.896]

N 1234 864 No of countries 60 59 Joint significance of election coefficients: P-value 0.007 0.002 m1:p-value 0.002 0.019 m2:p-value 0.341 0.114 Hansen OID: p-value: 0.808 0.132 No of instruments 17 18

Note: Windmeijer (2005) corrected standard errors in brackets. All equations are estimated using the two-step System-GMM with Windmeijer (2005) correction of standard errors. In column 2, the model controls for the self-selection bias associated with the participation into IMF programs through a two-step approach a la Maddala (1983), Vella and Verbeek (1999), and Keen and Lockwood (2010).

* p < 0.10,

** p < 0.05,

*** p < 0.01.